WO2023060632A1

WO2023060632A1 - Street view ground object multi-dimensional extraction method and system based on point cloud data

Info

Publication number: WO2023060632A1
Application number: PCT/CN2021/124565
Authority: WO
Inventors: 罗再谦; 刘颖; 向煜; 黄志�; 周兵; 韩�熙; 华媛媛; 朱勃; 李兵; 张彦; 曹欣; 王永刚; 王军涛; 李楠楠; 王翔
Original assignee: 重庆数字城市科技有限公司
Priority date: 2021-10-14
Filing date: 2021-10-19
Publication date: 2023-04-20
Also published as: CN113963259A

Abstract

A street view ground object multi-dimensional extraction method and system based on point cloud data. The method comprises: original point cloud data comprising absolute coordinates and elevation information; extracting a ground point cloud and a non-ground point cloud from the point cloud data by using a cloth algorithm; and extracting, from the non-ground point cloud, various ground objects type by type by using a multi-dimensional method, so as to achieve identification of different types of ground objects. Different types of ground objects are extracted from non-ground point cloud data by using a multi-dimensional method, such that street ground objects can be quickly, effectively and accurately extracted and identified.

Description

Method and system for multi-dimensional extraction of street scene features based on point cloud data

technical field

The present invention relates to the technical field of point cloud data extraction, in particular to a method and system for multi-dimensional extraction of street view features based on point cloud data.

Background technique

In recent years, laser radar technology (LIDAR) is a technology that can directly obtain three-dimensional information on the ground, which has unparalleled advantages over traditional aerial photogrammetry. The laser radar scanning system avoids the necessary steps of orientation and image matching in traditional photogrammetry. Compared with the traditional photogrammetry method, it has the advantages of fast speed, high precision, and low cost, and has the advantages of real-time, high efficiency, non-contact, and large data volume. The point cloud data acquired by LIDAR contains rich environmental information, including ground information, vegetation information, wire information, building information, etc. To accurately obtain the 3D information of buildings, the LIDAR data must be processed.

Due to the diversity and complexity of the geometric model of the building, the artificial environment and the natural environment in which the building is located are generally more complicated, including vegetation such as trees, as well as roads, poles and towers and other artificial features. Objects are extracted through a single dimension, and the types of extracted objects are single, and the accuracy is not high.

Contents of the invention

(1) Purpose of the invention

In view of the above problems, the purpose of the present invention is to propose a multi-dimensional extraction method and system for street scene features based on point cloud data, by using multi-dimensional methods to extract different types of features from non-ground point cloud data, which can be quickly and effectively 1. Accurately extracting street features, the present invention discloses the following technical solutions.

(2) Technical solution

As the first aspect of the present invention, the present invention discloses a multi-dimensional extraction method of street view features based on point cloud data, including:

Raw point cloud data includes absolute coordinates and elevation information;

Using a cloth algorithm to extract ground point clouds and non-ground point clouds in the point cloud data;

A multi-dimensional method is used to extract various ground objects by category from the non-ground point cloud to obtain different types of ground object recognition.

In a possible implementation manner, the multi-dimensional method is used to extract various ground objects by category from the non-ground point cloud, which specifically includes;

Using an elevation value and a watershed algorithm to extract tree class point cloud data in the non-ground point cloud;

Extracting rod-like object point cloud data in the non-ground point cloud through elevation values and hierarchical clustering;

Based on the elevation value and the multi-level semantics, the building point cloud data in the non-ground point cloud is extracted.

In a possible implementation manner, the extracting the tree class point cloud data in the non-ground point cloud specifically includes:

Carrying out grid division of the non-ground point cloud data, and projecting it into a grayscale image;

Carrying out watershed segmentation to the grayscale image to determine the outline of street trees;

The street tree point cloud is extracted from the non-ground point cloud data according to the street tree outline.

In a possible implementation manner, the extracting the rod-like object point cloud data in the non-ground point cloud specifically includes:

layering the non-ground point cloud according to preset elevation intervals;

Clustering the non-ground point cloud data of each layer;

According to the characteristics of the rod-shaped objects, the second clustering is performed on the layered clustering results in the elevation direction to extract the point cloud of the rod-shaped objects.

In a possible implementation manner, the extracting the building point cloud data in the non-ground point cloud specifically includes:

Extract point clouds of high-rise buildings above the threshold height through single-point semantic features;

Project the point cloud and high-rise building point cloud below the threshold to the XOY plane, and divide the grid according to the preset size, and select the grid of interest according to the semantic characteristics of the grid;

Connectivity analysis is performed on the interest grid to obtain the object area, and based on the regional semantic features, the building facade point cloud is extracted.

As the second aspect of the present invention, the present invention also discloses a multi-dimensional extraction system of street view features based on point cloud data, including:

A preprocessing module, which preprocesses the original point cloud data, the original point cloud data including absolute coordinates and elevation information;

An extraction module, the extraction module extracts ground point clouds and non-ground point clouds in the point cloud data by using a cloth algorithm;

A recognition module, the recognition module uses a multi-dimensional method to extract various ground objects by category from the non-ground point cloud, and obtains different types of ground object recognition.

In a possible implementation manner, the identification module includes identifying tree-like units, identifying rod-shaped object-like units, and identifying building-like units;

Described recognition tree class unit adopts elevation value and watershed algorithm, extracts the tree class point cloud data in described non-ground point cloud;

The identifying rod-shaped object class unit extracts the rod-shaped object point cloud data in the non-ground point cloud through the elevation value and hierarchical clustering;

The identification of building-type units extracts building-type point cloud data in the non-ground point cloud based on elevation values and multi-level semantics.

In a possible implementation manner, the identifying tree-like units includes dividing subunits, dividing subunits, and extracting tree-like subunits;

The division subunit divides the non-ground point cloud data into a grid and projects it into a grayscale image;

The segmentation subunit performs watershed segmentation on the grayscale image to determine the outline of street trees;

The extracting tree subunit extracts street tree point clouds from the non-ground point cloud data according to street tree outlines.

In a possible implementation manner, the unit for identifying rod-shaped objects includes a hierarchical subunit, a clustering subunit, and a subunit for extracting rod-shaped objects;

The layering subunit layers the non-ground point cloud according to a preset elevation interval;

The clustering subunit clusters the non-ground point cloud data of each layer;

The extracting rod-shaped feature subunit performs a second clustering on the layered clustering results in the elevation direction according to the characteristics of the rod-shaped feature, and extracts the point cloud of the rod-shaped feature.

In a possible implementation manner, the identifying building type unit includes a semantic subunit, a grid subunit and the extracted building facade subunit;

The semantic subunit extracts point clouds of high-rise buildings above a threshold height through single-point semantic features;

The grid subunit projects the point cloud and high-rise building point cloud below the threshold to the XOY plane, and divides the grid according to the preset size, and selects the grid of interest according to the semantic features of the grid;

The extracting building facade subunit performs connectivity analysis on the interest grid to obtain object areas, and extracts building facade point clouds based on regional semantic features.

(3) Beneficial effects

The present invention discloses a method and system for extracting multi-dimensional street scene features based on point cloud data, which has the following beneficial effects: the collected point cloud data is preprocessed, the ground point cloud and non-ground point cloud are extracted by using the cloth algorithm, By adopting multi-dimensional or multi-type methods to extract various ground object ranges from non-ground point clouds, the contours and categories of ground objects can be extracted and identified with high efficiency, high precision and accuracy.

Description of drawings

The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain and describe the present invention, but should not be construed as limiting the protection scope of the present invention.

Fig. 1 is one of the flow charts of the multi-dimensional extraction method of street view features based on point cloud data disclosed by the present invention;

Fig. 2 is the second flow chart of the multi-dimensional extraction method of street view features based on point cloud data disclosed by the present invention;

Fig. 3 is a flow chart of extracting tree class point cloud data in the non-ground point cloud disclosed by the present invention;

Fig. 4 is the elevation difference schematic diagram of different positions of the street tree point cloud model disclosed by the present invention;

Fig. 5 is the waveform diagram of the elevation difference distribution of the single street tree point cloud disclosed by the present invention;

Fig. 6 is a flow chart of extracting rod-shaped object point cloud data in the non-ground point cloud disclosed by the present invention;

Fig. 7 is a flow chart of extracting building class point cloud data in the non-ground point cloud disclosed by the present invention;

Fig. 8 is a block diagram of a multi-dimensional extraction system for street view features based on point cloud data disclosed in the present invention.

Reference numerals: 400, preprocessing module; 500, extraction module; 600, recognition module; 610, identification of tree-like units; 611, division of sub-units; 612, segmentation of sub-units; 613, extraction of tree-like sub-units; 620, identification Rod-shaped object unit; 621, hierarchical subunit; 622, clustering subunit; 623, extracting rod-shaped object subunit; 630, identifying building class unit; 631, semantic subunit; 632, grid subunit ; 633. Extract building facade sub-units.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention will be described in more detail below in conjunction with the drawings in the embodiments of the present invention.

It should be noted that: in the drawings, the same or similar symbols represent the same or similar elements or elements with the same or similar functions. The described embodiments are part of the embodiments of the present invention, but not all of the embodiments. In the case of no conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

In describing the present invention, it is to be understood that the terms "central", "longitudinal", "transverse", "front", "rear", "left", "right", "vertical", "horizontal", The orientations or positional relationships indicated by "top", "bottom", "inner", "outer", etc. are based on the orientations or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying The devices or elements referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore should not be construed as limiting the scope of the invention.

A first embodiment of a multi-dimensional extraction method of street view features based on point cloud data disclosed in the present invention will be described in detail below with reference to FIGS. 1-7 . This embodiment is mainly applied to point cloud data extraction. By using a multi-dimensional method to extract different types of ground features from non-ground point cloud data, street features can be extracted quickly, effectively and accurately.

As shown in Figure 1-2, this embodiment specifically includes the following steps:

S100. The original point cloud data includes absolute coordinates and elevation information.

In step S100, in this application, the vehicle-mounted mobile measurement system is used to collect the original point cloud data. Specifically, the vehicle-mounted laser scanning system is used to acquire data including laser point cloud data. The laser point cloud data is a set of points with three-dimensional coordinates in real space.

Further, after the original point cloud data is collected, the original point cloud data is preprocessed, and the original point cloud data is preprocessed by denoising, filtering and other technologies, so as to improve the accuracy of the point cloud data.

Furthermore, the vehicle-mounted laser scanning system adopts a vehicle-mounted mobile data acquisition system with various advanced technologies such as GPS positioning, INS inertial navigation, CCD video, and automatic control. It uses vehicle-mounted remote sensing to quickly collect street and road positioning information on the spot. There are two methods of measurable stereoscopic images and using professional cameras to collect single-point panoramic images and 360° panoramic images. The collected data has the advantages of simple and fast update, high efficiency, and rich information.

S200. Using a cloth algorithm, extract ground point clouds and non-ground point clouds in the point cloud data.

In step S200, the cloth algorithm is used to extract the ground point cloud and non-ground point cloud in the point cloud data, specifically including:

S210. Initialize the cloth grid, and determine the number of grid nodes through the grid resolution;

S220. Project the point cloud data and the grid points onto the same horizontal plane, determine the point cloud data corresponding to each grid point, and mark the elevation value of the corresponding point cloud data point;

S230. Calculate the position of the grid node moved by gravity, and compare the elevation of the position with the elevation of the corresponding point cloud data point. If the node elevation is less than or equal to the point cloud data elevation, replace the node position with the position of the corresponding point cloud data point, and mark it as an immovable point. The position of the cloth point after being displaced by gravity is calculated by the following formula:

Where X is the position of the cloth grid point at time t, Δt is the time period, G is the acceleration and is a constant value, and A is the quality of the cloth grid point, which is set as a constant 1.

S240. Calculate the moving position of each grid point affected by the neighboring nodes.

S250, repeating steps S230 and S240, when the maximum elevation change of all nodes is small enough or exceeds the maximum number of iterations, the simulation process is terminated;

S250. Classify ground point clouds and non-ground point clouds, and calculate distances between grid points and corresponding point cloud data points. For point cloud data points, if the distance is less than the threshold L, it is classified as ground point cloud, otherwise it is classified as non-ground point cloud.

S300. Using a multi-dimensional method to extract various ground objects one by one from the non-ground point cloud, and obtain different types of ground object recognition.

In step S300, a multi-dimensional method is used to extract various ground objects one by one from the non-ground point cloud, specifically including;

S310. Using the elevation value and the watershed algorithm, extract tree-like point cloud data in the non-ground point cloud.

As shown in Figure 3, in step S310, the extraction of trees, that is, the extraction of street trees, specifically includes:

S311. Grid-dividing the non-ground point cloud data and projecting it into a grayscale image;

S312. Carry out watershed segmentation on the grayscale image, and determine the outline of street trees;

S313. Extract street tree point clouds from non-ground point cloud data according to street tree outlines.

The non-ground point cloud data is grid-divided and projected into a grayscale image, and the gray-scale image is divided into watershed to determine the outline of the street tree, and then the complete street tree point cloud is extracted from the non-ground point cloud data according to the outline of the street tree. Among them, the elevation difference of different positions of the non-ground point cloud is obtained through the elevation model, and the gray image is segmented according to the elevation difference of different positions of the non-ground point cloud to determine the outline of the street tree. For example, the elevation difference distribution of a single street tree is shown in Figure 4. The elevation difference of different positions of the street tree point cloud model is shown. As shown in Figure 5, it shows the waveform diagram of the elevation difference distribution of a single street tree point cloud. The horizontal axis is the projection diameter D of a single street tree crown on the XOY plane, and the vertical axis is is the elevation difference. It can be seen that the closer to the crown center or trunk, the greater the elevation difference, and the closer to the peak, and the farther away from the crown center or trunk, the smaller the elevation difference, and the closer to the trough.

S320. Using the elevation value and hierarchical clustering, extract point cloud data of rod-like objects in the non-ground point cloud.

As shown in Figure 6, in step S320, extract the point cloud data of rod-shaped objects in the non-ground point cloud, specifically including:

S321. Layer the non-ground point cloud according to the preset elevation interval;

S322. Clustering the non-ground point cloud data of each layer;

S323. Perform a second clustering on the layered clustering results in the elevation direction according to the characteristics of the rod-shaped objects to extract point clouds of the rod-shaped objects.

The non-ground point cloud is layered according to the preset elevation interval, and the point cloud data of each layer is clustered, and then according to the characteristics of the continuous extension of the rod-shaped objects in the elevation direction, the layered clustering results are reassessed in the elevation direction. Clustering to identify the complete rod-shaped feature point cloud.

Furthermore, this application uses the method of processing data from high to low to layer the non-ground point cloud according to the preset elevation interval, first determine the maximum and minimum values of the non-ground point cloud in the elevation direction, and divide the obtained results in the elevation direction The number of non-ground point cloud layers, the threshold value in the elevation direction of each layer, and the distance between each layer are set, and the non-ground point cloud is clustered according to the above rules. In the elevation direction, the data processing process is adopted from high to low, without data filtering, which simplifies the data processing process, involves fewer parameters, data processing is efficient, and the degree of automation is higher.

S330. Based on the elevation value and multi-level semantics, extract point cloud data of buildings in the non-ground point cloud.

As shown in Figure 7, in step S330, the building class point cloud data in the non-ground point cloud is extracted, specifically including:

S331. Extracting point clouds of high-rise buildings above a threshold height through single-point semantic features;

S332. Project the point cloud below the threshold and the point cloud of high-rise buildings to the XOY plane, divide the grid according to the preset size, and select the grid of interest according to the semantic characteristics of the grid;

S333. Perform connectivity analysis on the interest grid to obtain the object area, and extract building facade point clouds based on the semantic features of the area.

Firstly, through the single-point semantic feature, that is, the elevation value of the point, the point cloud lower than the building is eliminated, and the high-rise building point cloud containing only the building above a certain height is extracted at the same time; then the remaining point cloud and the high-rise building point cloud are projected to XOY Plane and divide the grid according to a certain size, and select the grid of interest according to the semantic characteristics of the grid; finally, analyze the connectivity of the grid of interest to obtain the object area, and realize the accurate extraction of building facade point clouds based on the semantic features of the area.

Further, analyze the connectivity of the grid of interest, specifically whether there is a projection point cloud connection between each grid. For example, the projection point cloud in some grids is in a small piece in the middle, and there is a gap with the grid boundary. Then this grid is not connected to the surrounding grids, otherwise, this grid is connected to the surrounding grids.

Further, when using a multi-dimensional or multi-category method to extract feature point cloud data, each dimension extracts/recognizes a type of feature, wherein the range of various features can be extracted by category, and the extracted feature range For fusion, various ground object ranges can also be extracted at the same time to complete the ground object recognition.

The following describes in detail with reference to FIG. 4-FIG. 5 and FIG. 8. Based on the same inventive concept, the embodiment of the present invention also provides a first embodiment of a multi-dimensional extraction system of street view features based on point cloud data. Since the principle of the problem solved by this system is similar to the aforementioned multi-dimensional extraction method of street view features based on point cloud data, the implementation of this system can refer to the implementation of the aforementioned method, and the repetition will not be repeated.

This embodiment is mainly applied to cloud data extraction. By using a multi-dimensional method to extract different types of ground features from non-ground point cloud data, street features can be extracted quickly, effectively and accurately.

As shown in FIG. 8 , this embodiment mainly includes: a preprocessing module 400 , an extraction module 500 and an identification module 600 .

Wherein, the preprocessing module 400 preprocesses the original point cloud data, and the original point cloud data includes absolute coordinates and elevation information; the extraction module 500 uses the cloth algorithm to extract the ground point cloud and non-ground point cloud in the point cloud data; the identification module For non-ground point clouds, use multi-dimensional methods to extract various ground objects one by one, and obtain different types of ground object recognition.

Further, in this application, the vehicle-mounted mobile measurement system is used to collect the original point cloud data, and the vehicle-mounted laser scanning system is used to acquire data including laser point cloud data. The laser point cloud data is a set of points with three-dimensional coordinates in real space.

Further, after the original point cloud data is collected, the original point cloud data can be preprocessed, and the original point cloud data can be preprocessed by denoising, filtering and other technologies, thereby improving the accuracy of the point cloud data.

In a possible implementation manner, the extraction module 500 uses a cloth algorithm to extract ground point clouds and non-ground point clouds in the point cloud data, specifically: initializing the cloth grid, and determining the number of grid nodes through the grid resolution; Project the point cloud data and grid points to the same horizontal plane, determine the point cloud data corresponding to each grid point, and mark the elevation value of the corresponding point cloud data point; calculate the position of the grid node moved by gravity, and compare the position elevation It corresponds to point cloud data point elevation. If the node elevation is less than or equal to the point cloud data elevation, replace the node position with the position of the corresponding point cloud data point, and mark it as an immovable point. The position of the cloth point after being displaced by gravity is calculated by the following formula:

Where X is the position of the cloth grid point at time t, Δt is the time period, G is the acceleration and is a constant value, A is the quality of the cloth grid point, which is set to a constant 1; calculate the influence of each grid point by the adjacent nodes and the moving position; repeat the above calculation of the position of grid nodes moved by gravity and calculate the position of each grid point moved by the influence of adjacent nodes. When the maximum elevation change of all nodes is small enough or exceeds the maximum number of iterations, the simulation process Terminate; classify the ground point cloud and non-ground point cloud, and calculate the distance between the grid points and the corresponding point cloud data points. For point cloud data points, if the distance is less than the threshold L, it is classified as a ground point cloud, otherwise it is classified as a non-ground point cloud.

In a possible implementation, the recognition module 600 includes a tree recognition unit 610, a rod-shaped object recognition unit 620, and a building recognition unit 630, wherein the tree recognition unit 610 uses elevation values and watershed algorithms to extract non-ground Tree-like point cloud data in the point cloud; identify pole-shaped object unit 620 through elevation values and hierarchical clustering to extract pole-shaped object-type point cloud data in non-ground point clouds; identify building class unit 630 based on elevation values and multi-level semantics to extract building-like point cloud data from non-ground point clouds.

In a possible implementation, the identifying tree class unit 610 includes a division subunit 611, a division subunit 612, and an extraction tree class subunit 613; wherein the division subunit 611 divides the non-ground point cloud data into grids, and projects is a grayscale image; the segmentation subunit 612 performs watershed segmentation on the grayscale image to determine the outline of the street tree; the extracting tree class subunit 613 extracts the street tree point cloud from the non-ground point cloud data according to the outline of the street tree.

In a possible implementation, the unit for identifying rod-shaped objects 620 includes a layering subunit 621, a clustering subunit 622, and a subunit for extracting rod-shaped objects 623, wherein the layering subunit 621 treats non-ground point clouds according to The preset elevation interval is stratified; the clustering subunit 622 clusters the non-ground point cloud data of each layer; The clustering results are clustered for the second time, and the rod-shaped object point cloud is extracted.

In a possible implementation manner, the recognition building class unit 630 includes a semantic subunit 631, a grid subunit 632, and an extraction building facade subunit 633, wherein the semantic subunit 631 uses single-point semantic features to extract The high-rise building point cloud; the grid subunit 632 projects the point cloud below the threshold and the high-rise building point cloud to the XOY plane, and divides the grid according to the preset size, and selects the grid of interest according to the semantic features of the grid; extracts the building vertical The face subunit 633 performs connectivity analysis on the interest grid to obtain the object area, and extracts the building facade point cloud based on the semantic features of the area.

The above is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention shall be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

A multi-dimensional extraction method for street scene features based on point cloud data, characterized in that it includes:

Raw point cloud data includes absolute coordinates and elevation information;

Using a cloth algorithm to extract ground point clouds and non-ground point clouds in the point cloud data;

A multi-dimensional method is used to extract various ground objects by category from the non-ground point cloud to obtain different types of ground object recognition.
The multi-dimensional extraction method of street view features based on point cloud data according to claim 1, wherein the non-ground point cloud is extracted by using a multi-dimensional method to extract various features by category, specifically comprising;

Using an elevation value and a watershed algorithm to extract tree class point cloud data in the non-ground point cloud;

Extracting rod-like object point cloud data in the non-ground point cloud through elevation values and hierarchical clustering;

Based on the elevation value and the multi-level semantics, the building point cloud data in the non-ground point cloud is extracted.
The multi-dimensional extraction method of street scene features based on point cloud data according to claim 2, wherein said extraction of tree class point cloud data in said non-ground point cloud specifically includes:

Carrying out grid division of the non-ground point cloud data, and projecting it into a grayscale image;

Carrying out watershed segmentation to the grayscale image to determine the outline of street trees;

The street tree point cloud is extracted from the non-ground point cloud data according to the street tree outline.
The multi-dimensional extraction method of street scene feature based on point cloud data according to claim 2, is characterized in that, described extracting the pole-shaped object class point cloud data in described non-ground point cloud specifically comprises:

layering the non-ground point cloud according to preset elevation intervals;

Clustering the non-ground point cloud data of each layer;

According to the characteristics of the rod-shaped objects, the second clustering is performed on the layered clustering results in the elevation direction to extract the point cloud of the rod-shaped objects.
The multi-dimensional extraction method of street scene feature based on point cloud data according to claim 2, wherein said extracting the building class point cloud data in said non-ground point cloud specifically includes:

Extract point clouds of high-rise buildings above the threshold height through single-point semantic features;

Project the point cloud and high-rise building point cloud below the threshold to the XOY plane, and divide the grid according to the preset size, and select the grid of interest according to the semantic characteristics of the grid;

Connectivity analysis is performed on the interest grid to obtain the object area, and based on the regional semantic features, the building facade point cloud is extracted.
A multi-dimensional extraction system of street scene features based on point cloud data, characterized in that it includes:

A preprocessing module, which preprocesses the original point cloud data, the original point cloud data including absolute coordinates and elevation information;

An extraction module, the extraction module extracts ground point clouds and non-ground point clouds in the point cloud data by using a cloth algorithm;

A recognition module, the recognition module uses a multi-dimensional method to extract various ground objects by category from the non-ground point cloud, and obtains different types of ground object recognition.
The multi-dimensional extraction system of street view features based on point cloud data according to claim 6, wherein the identification module includes identifying tree-like units, identifying rod-shaped object-like units and identifying building-like units;

The recognition tree class unit adopts the elevation value and the watershed algorithm to extract the tree class point cloud data in the non-ground point cloud;

The identifying rod-shaped object class unit extracts the rod-shaped object point cloud data in the non-ground point cloud through the elevation value and hierarchical clustering;

The identification of building-type units extracts building-type point cloud data in the non-ground point cloud based on elevation values and multi-level semantics.
The multi-dimensional extraction system of street view features based on point cloud data according to claim 7, wherein said identifying tree class units includes dividing subunits, segmenting subunits and extracting tree class subunits;

The division subunit divides the non-ground point cloud data into a grid and projects it into a grayscale image;

The segmentation subunit performs watershed segmentation on the grayscale image to determine the outline of street trees;

The extracting tree subunit extracts street tree point clouds from the non-ground point cloud data according to street tree outlines.
The multi-dimensional extraction system of street view features based on point cloud data according to claim 7, wherein the unit for identifying rod-shaped objects includes a hierarchical subunit, a clustering subunit, and a subunit for extracting rod-shaped objects ;

The stratification subunit stratifies the non-ground point cloud according to a preset elevation interval;

The clustering subunit clusters the non-ground point cloud data of each layer;

The extracting rod-shaped feature subunit performs a second clustering on the layered clustering results in the elevation direction according to the characteristics of the rod-shaped feature, and extracts the point cloud of the rod-shaped feature.
The multi-dimensional extraction system of street view features based on point cloud data according to claim 7, wherein the recognition building class unit includes a semantic subunit, a grid subunit and an extraction building facade subunit;

The semantic subunit extracts point clouds of high-rise buildings above a threshold height through single-point semantic features;

The grid subunit projects the point cloud and high-rise building point cloud below the threshold to the XOY plane, and divides the grid according to the preset size, and selects the grid of interest according to the semantic features of the grid;

The extracting building facade subunit performs connectivity analysis on the interest grid to obtain object areas, and extracts building facade point clouds based on regional semantic features.